Please see my attachment with screenshots(multi-homed application servers network configuration.docx and others) in my ftp site, thanks for your help.
Hi Isaias Freitas,
we run multi homed hosts for a long time before. However, I did not see the note SAP mentioned. We ran our SAP instances in server network but we found RFC load balancing returned server network IP address(which is unreachable from access network) recently.
Because I implement this network architecture for a long time and I believe it works. However, after reading the SAP note of multi homed host, I found I have to run SAP in access network, right?
I see SAP recommend use static route but it still has some issue with cluster. I develop my own network architecture, that is, running SAP host in server network(a SAP domain in server network, too) and use access/public network at first one in the binding order.
However, this will make SAP message server connect to other application servers using access network because SAP instances are running in access network, that is the Gigabit network. So I observe SM66 is a little slower than before. I am not sure if this is correct. Could I let SAP instances communicate using 10Gb network? However, becuase my domain is in SAP server network, SAP message server should communicate using server network, right? Now I have no ideas why SM66 is much slow right now.
Yes, the message server will connect to the instances using the public network, that is the expected behavior. And this should not cause any harm, it's actually the opposite: the first NIC in the binding order must be the NIC that will be reachable by the users, otherwise it can cause some functions to not work.
You didn't mention which SAP note you have read, but you can refer to the SAP notes below:
21151 - Multiple Network adapters in SAP Servers
1033987 - Remote login using NAT or SAP router fails (you can read this one even if there is no NAT or saprouter involved)
Note that the NIC binding order must be correct before you start the SAP instance. If you've changed the binding order while the SAP was running, you need to restart the particular instance.
Regarding the SM66 being slower, SM66 uses RFC's to gather the information from the other/remote instances, thus it will depend on your network performance. If I've understood your last reply, the "access network" is the public network, through which the users can reach the SAP, and it is a Gigabit network, and the other (internal) network is a 10Gigabit network. Maybe that's why it felt a little slower.
Please reply back if I can further help you!
Hi Isaias Freitas,
I do appreciate your very detailed reply.
Note 21151 is the note I read.
The weird part is I have run SAP instances on server(inetrnal, 10Gb) network for serveral years becuase I want SAP always communicate each other using 10Gb NIC. And I believe(maybe I were wrong) SAP servers know how to communicate with SAP server or SAP frontend. If the host comes from server network, it communicates uing server network(192.168.28.X, netmask:255.255.255.0). If the hosts come from access(public) network, it communicates using access network(10.243.23.X, netmask: 255.255.255.0).
By the way, the attachment in that note mentioned that maybe it will work but SAP can't support it like the situation I mention and maybe it will go wrong after upgrading. Is this why I encounter this error?
Anyway, one day, a SAP consultant recommended me to upgrade kernel form 7.00 to 7.20 and I found sunddenly RFC load balancing sometimes failed becuase it got a server network IP address and it was unreachable. I found everthing went wrong so I read the note and the attachment and I configured the access(public) network at the first in the binding order.
I had read your reply and the Tcode:SM66 is a little slow is normal, right? Becuase now all SAP instances communicate(frontend and RFC logon blancing, SM66, AL08...etc) using public network(1Gb).
You're welcome :-).
Yes, the SAP system will be able to function if you change the binding order, making the 10Gb network the first one. However, some functions of the SAP will not work correctly. The RFC load balancing, that you've mentioned, is one of them. Another example is the "remote logon" button at SM59 (e.g., you're logged on to SAP system ABC, trying to perform a "remote logon" to SAP system XYZ, and ABC has the wrong network binding order; the "remote logon" from ABC to XYZ will fail).
Regarding the kernel, the kernel 720 will substitute older kernel releases. SAP note 1629598 has further details.
please follow 1st link to configure multi-homing
[Multihoming a Windows Server|http://networkadminkb.com/KB/a14/multihoming-a-windows-server.aspx]